A degradation evaluation method with the convolutional neural network for the cyclic symmetry rolling bearing

Abstract The rolling bearing is a cyclic symmetry engineering component, whose failure is related to the regular operation of the rotating machine. For the purpose of real-time monitoring of operational status and predicting the occurrence of the failure, a rolling bearing performance degradation assessment method based on the convolutional neural network (CNN) was proposed with the XJTU-SY tested acceleration life data set. Firstly, the fault signal characteristics of the rolling bearing were extracted by using an energy decomposition method with a wavelet package to form some data sets. Secondly, the degradation performance of the data is classified by the Fuzzy C-means Algorithm for fuzzy data, and then the extracted characteristic signals were processed with CNNs. Thirdly, a multi-layer nonlinear mapping of training data was firstly formed with CNN, and then the adaptive batch normalization adaptive network was added during the training process, and then the deep learning model was formed as a result. Finally, the degradation degree of the three typical rolling bearing failures was evaluated and predicted with this method. The t-SNE technology was employed to visualize the output of the data distribution in two-dimensional space. Compared with the experiment results, the accuracy of the outer ring, inner ring and rolling body of the rolling bearing are 90.84%, 83.3%, and 80.95% with the fault assessment method, respectively. Thus, the degradation of the cyclic symmetry rolling bearing can be identified effectively and determined by the proposed method in advance.